Causality-based multi-model ensemble learning for safety assessment in metro tunnel construction

•A causality-based multi-model ensemble learning approach is proposed for the safety assessment of MTC.•The causality of a data pair is proposed to reflect the causal input-output relation of a data pair in MTC.•All data pairs are divided into high/low-causality data with a causal/adverse-causal input-output relation.•Multiple sub-models are constructed using datasets according to their causality.•Outputs from the sub-models are integrated with consideration of the sub-model accuracy and the matching degree. The safety of the nearby buildings to the metro lines is directly affected by the underground metro tunnel construction (MTC) activities. In this study, a new causality-based multi-model ensemble learning approach is proposed for the safety assessment of MTC. First, data causality is defined to reflect the causal relation between the assessment input and output, and it is calculated using an improved ensemble learning approach. Then, multiple sub-models are constructed using different sub-datasets which are classified according to the data causality. Third, the weights of sub-multiple models are calculated according to the respective accuracy of the sub-models and the matching degrees between the new input and different sub-datasets. Finally, a unified output is obtained by integrating the outputs from sub-models while considering their respective weights. A practical case of building tilt rate (BTR) assessment of Metro Line 6 in the city of Wuhan, China, is studied. Case study results show that the proposed approach outperforms (1) using a single sub-model and several other machine learning approaches, and also (2) not adopting the data causality to classify sub-datasets. Moreover, how varied settings of the sub-datasets classification ratios and weight thresholds would affect the performance is also investigated.